Smart assignment
Smart Assignment routes conversations to human agents based on the customer's detected sentiment, so that upset, aggressive, or stuck customers reach a person sooner instead of staying with the AI. Use it to make sure the situations that most need a human — a customer asking for an agent, a customer getting angry, a conversation the AI can't resolve — are prioritised in the queue automatically.
The page has four sections:
- Assignment Priority — one row per sentiment scenario the AI watches for.
- SLA Breach Priority — an extra boost for tickets that breach an SLA level.
- Repeat Contact Priority — an extra boost for tickets whose customer keeps reaching out. See Repeat contact priority.
- Reopened Ticket Priority — a positive or negative adjustment for tickets with a recorded reopen count.
Edits in all four are staged; click Save Changes at the bottom to apply them.
Assignment priority
For each scenario you configure up to three things:
- Status — a toggle that enables or disables detection for that scenario. When off, the AI ignores that sentiment for routing.
- Priority — a number from 0 to 10 that sets how urgently a matching ticket is handled. Higher numbers are handled first. Editable only while the scenario is enabled.
- Auto-create ticket — a toggle that creates a support ticket for a human as soon as the scenario is detected, rather than waiting for the normal flow. Editable only while the scenario is enabled.
Scenarios
Ordered as they appear in the table, roughly from mild to severe:
- Asking for Agent — the customer requests human assistance ("Can I speak to someone?", "Connect me to support").
- Urgently Requesting Agent — the customer urgently or repeatedly demands a human ("I need to speak to someone NOW").
- Frustrated Customer — early signs of dissatisfaction ("This is not working", "Why is this so difficult?").
- Highly Frustrated Customer — extremely upset or threatening to leave ("This is unacceptable!", "I want to cancel", profanity or repeated caps).
- Aggressive Customer — hostile language or threats ("You people are thieves", "This is a scam").
- Highly Aggressive Customer — abusive language, serious or legal threats, extreme verbal abuse.
- Potential Escalation — the customer threatens to take a complaint to legal authorities, consumer court, company leadership, social media, public reviews, or the press. Calm and conditional threats count; neutral mentions and routine supervisor requests do not. Disabled by default, with a suggested priority of 3 and auto-create ticket off.
- Unable to Resolve — the AI has hit its automation limit for the conversation and can't resolve it on its own, or it can't use a customer's answer (see below). This scenario is Always on (its Status can't be turned off) and defaults to auto-creating a ticket, so a stuck conversation always ends up with a human. You can still adjust its priority and its auto-create toggle.
- Disputed Outcome — the customer doesn't accept the outcome of a request the AI resolved, for example a refusal under your policy. It applies only after a task completes with its completion message or closes through a closing branch (a transition that ends the task without moving to another task). On by default. See Disputed outcomes.
Disputed outcomes
When the AI resolves a request and the customer replies that they don't accept it (they disagree, call it unfair, or insist on what was refused), the AI counts a dispute:
- Before the handoff count, the AI acknowledges the concern and explains again why the outcome stands, using the reason it gave and your policy. It doesn't reverse the outcome or promise an agent.
- On the dispute set in Hand off on dispute (default 2), the AI tells the customer an agent will review the request, and the resolved ticket reopens to the queue tagged Disputed Outcome. Set it to 1 to hand off on the first dispute, or 0 to never hand off. With Auto-create ticket off, the AI never hands off.
- A customer who brings new facts the outcome didn't consider, such as describing damage or sending photos, isn't counted as disputing. Their message is handled normally.
- Count picking a different issue for the same item as a dispute (off by default) also counts a customer who, right after the outcome, picks another issue type for the same item without saying what actually happened differently.
Counts last for the chat session. A task, or one closing branch within it, can set its own dispute count and handoff team in the task editor, and that overrides this row. An explicit request for an agent is still handled by Asking for Agent.
When the AI can't use a customer's answer
In these cases the AI stops asking and hands the conversation over through Unable to Resolve, using its priority. Unable to Resolve is always on, so nothing needs to be set up:
- Your order system isn't responding. When an order lookup is rate limited, times out, or returns a server error, the AI retries briefly. If it still fails, the AI keeps the order number the customer gave. It doesn't call the order invalid or ask for another one. It tells the customer it can't pull up the details right now and, when Asking for Agent creates tickets, offers to connect them to the team. If the lookup fails again on their next message, the AI hands over. The ticket names the order and what failed.
- A required answer keeps being rejected. After the objective's Max asks (default 3), the AI hands over instead of asking again. The ticket names the objective and the last answer. See When an answer doesn't settle an objective.
- The customer can't provide a required answer, for example they don't have the invoice. The AI hands over instead of asking again.
With Unable to Resolve's Auto-create ticket on (the default), a ticket is created and the AI tells the customer the team will get back to them. With it off, no ticket is created: the AI tells the customer it can't complete the request in chat and that they can reach the support team directly. Either way it stops asking for that answer.
SLA breach priority
Each SLA configuration can add a boost of 0 to 10 for each level (L1, L2, L3). When a ticket breaches a level, that level's boost is added to its priority. Levels add up, so a ticket that reaches L3 gets the L1 + L2 + L3 boosts, and boosts from different SLAs add up too. Thresholds are managed in SLA settings.
Repeat contact priority
Moves a ticket up the queue when the customer keeps reaching out about it. Off by default; turn it on with the switch at the top of the section.
Contacts are counted across the ticket, its child tickets, and its interaction tickets, by channel:
- Chat — one per channel per day on which the customer sent a WhatsApp, Instagram, or live chat message.
- Email — one per email thread the customer wrote in.
- Calls — one per incoming call.
Only contacts where the customer wrote or called in count. Agent outreach, template messages, and outbound calls (including AI calls) don't.
Each channel has a Status toggle and up to three tiers. A tier reads "N or more interactions → boost" (N from 1 to 100, boost 0 to 10); clear the count to leave a tier unused. The highest tier the count reaches applies. The boosts of the three channels add up, capped at 10.
For example, with Chat tier "3 or more → +2" and Calls tier "2 or more → +2", a customer who chatted on 3 different days and called twice adds +4 to the ticket's priority.
The boost updates whenever a new contact is recorded. Saving a change re-scores open tickets, and turning the section off resets their boost to 0. Tickets with a boost show a Repeat +N tag.
Reopened ticket priority
Set Priority adjustment from −10 to +10 to change queue priority for tickets whose Reopen Count is greater than zero. The default is 0, which leaves priority unchanged. Negative values lower priority; positive values raise it.
The adjustment applies once, regardless of the count: −2 subtracts two priority points whether the ticket reopened once or five times. A child ticket qualifies when either its own or its parent's reopen count is greater than zero; it still receives only one adjustment.
This uses the same count as the Reopened ticket filter. Customer replies during resolution confirmation count, including email replies while automatic resolution is pending. Manual reopening, ordinary awaiting-response replies, and snooze expiry do not increment this count. Once a ticket has a positive count, the adjustment applies whenever it is queued for automatic assignment.
Saving changes affects subsequent assignment scoring for queued tickets. SLA and repeat contact boosts still apply. Queue aging also raises the score based on the ticket's original creation time, so a negative adjustment does not guarantee that reopened tickets come after new tickets. The previous-agent routing and capacity rules in Agent Assignement & Team still apply.
How it works
As the AI reads incoming messages it classifies the customer's sentiment against these scenarios. When an enabled scenario matches, the conversation is tagged with that scenario's priority value, and — if auto-create ticket is on — a ticket is created for a human right away.
A ticket's priority is its sentiment priority plus any SLA breach and repeat contact boosts and its reopened ticket adjustment. Because higher priority is handled first, the ordering of your priority values is what actually decides who jumps the queue. A common setup is to give the "aggressive" and "urgent" scenarios higher priority than the milder "frustrated" or "asking for agent" ones, so the most volatile situations are picked up first. Keep Unable to Resolve enabled (it always is) so conversations the AI genuinely can't handle don't get stuck.
Which agent actually receives the routed ticket, and the capacity and team rules that apply, are governed elsewhere — see Agent Assignement & Team. Smart Assignment decides when and how urgently a conversation should reach a human; Agent Assignement & Team decides who.
Sentiment updates for linked Sprinklr cases
When Flowcall automatically updates a linked Sprinklr case, it retains the strongest detected sentiment across the Flowcall tickets linked to that case. A calmer follow-up does not automatically lower an earlier Potential Escalation / P0 to Frustrated User / P3. A follow-up with no detected sentiment also preserves the earlier severity. This applies to future automatic updates; existing cases are not automatically reprocessed.
Related
- Agent Assignement & Team — how tickets are distributed to agents and teams once they're routed to humans.